The development of intrinsic capacity measures for longitudinal research : The Longitudinal Aging Study Amsterdam

This is a self-archived version of an original article. This version may differ from the original in pagination and typographic details. Author(s): Title: Year: Version: Copyright: Rights: Rights url: Please cite the original version: CC BY 4.0 https://creativecommons.org/licenses/by/4.0/ The development of intrinsic capacity measures for longitudinal research : The Longitudinal Aging Study Amsterdam © 2024 The Authors. Published by Elsevier Inc. Published version Qi, Yuwei; Schaap, Laura A.; Schalet, Benjamin D.; Hoogendijk, Emiel O.; Deeg, Dorly J.H.; Visser, Marjolein; Koivunen, Kaisa; Huisman, Martijn; van Schoor, Natasja M. Qi, Y., Schaap, L. A., Schalet, B. D., Hoogendijk, E. O., Deeg, D. J., Visser, M., Koivunen, K., Huisman, M., & van Schoor, N. M. (2024). The development of intrinsic capacity measures for longitudinal research : The Longitudinal Aging Study Amsterdam. Experimental Gerontology, 197, Article 112599. https://doi.org/10.1016/j.exger.2024.112599 2024 The development of intrinsic capacity measures for longitudinal research: The Longitudinal Aging Study Amsterdam Yuwei Qi a,* , Laura A. Schaap b,c , Benjamin D. Schalet e,f , Emiel O. Hoogendijk a,b,g , Dorly J.H. Deeg a , Marjolein Visser d , Kaisa Koivunen h , Martijn Huisman a,b,i , Natasja M. van Schoor a,b a Amsterdam UMC location Vrije Universiteit Amsterdam, Epidemiology and Data Science, the Netherlands b Aging and Later Life, Amsterdam Public Health Research Institute, the Netherlands c Department of Health Sciences, Faculty of Science, Amsterdam Public Health Research Institute, Amsterdam Movement Sciences, Vrije Universiteit Amsterdam, the Netherlands d Department of Health Sciences, Faculty of Science, Amsterdam Public Health Research Institute, Vrije Universiteit Amsterdam, the Netherlands e Amsterdam University Medical Centers, Department of Epidemiology and Data Science, the Netherlands f Amsterdam Public Health, Methodology, Amsterdam, The Netherlands g Department of General Practice, Amsterdam UMC Location Vrije Universiteit Amsterdam, the Netherlands h Faculty of Sport and Health Sciences and Gerontology Research Center, University of Jyv¨ askyl¨ a, Jyv¨ askyl¨ a, Finland i Department of Sociology, Vrije Universiteit Amsterdam, the Netherlands ARTICLE INFO Section Editor: Michael Drey Keywords: Intrinsic capacity Longitudinal measures Ageing Formative model ABSTRACT Background: The World Health Organization has introduced the construct of intrinsic capacity (IC) as an important component of healthy ageing and overall well-being in older adults The present study aimed to develop domain-specific and composite IC scores and to validate these scores by examining their longitudinal relation with functioning. Methods: We used prospective data on participants aged 57 to over 90 years, with a 10-year follow-up, from the Longitudinal Aging Study Amsterdam, an ongoing cohort study of older Dutch men and women Using a formative, stepwise approach, we identified indicators across the different domains of IC, i.e. vitality, sensory, cognition, psychology, and locomotion, using a combination of unidimensional factor analyses and Partial Least Squares Structural Equation Modelling (PLS-SEM). Next, domain-specific and composite IC scores were generated, and the construct validity (score across age groups) and criterion validity (relationship with change in functional limitations) were assessed. Results: The multiple unidimensional factor analyses and PLS-SEM identified a total of 18 indicators, covering the five domains of IC. The mean composite IC score was 70.9 (SD =0.9) in men and 69.7 (0.8) in women. The domain-specific and composite IC scores all showed good construct validity, with known-group validation results indicating age-related declines. A higher composite IC score was associated with less functional limitations over time (B =0.20, 95%CI [0.19, 0.22]). Conclusion: The developed domain-specific IC scores and the composite IC score effectively discriminated agerelated declines in IC. Additionally, the composite IC score was longitudinally associated with functional limitations. By creating this comprehensive and reliable tool for tracking IC, we aim to provide valuable insights into the dynamics of ageing and support more effective strategies for promoting health and well-being throughout later life. These scores establish a foundation for future research to track longitudinal changes across various IC domains and relate these changes to key age-related outcomes. * Corresponding author. E-mail address: [email protected] (Y. Qi). Contents lists available at ScienceDirect Experimental Gerontology journal homepage: www.elsevier.com/locate/expgero https://doi.org/10.1016/j.exger.2024.112599 Received 2 August 2024; Received in revised form 13 September 2024; Accepted 26 September 2024 Experimental Gerontology 197 (2024) 112599 Available online 9 October 2024 0531-5565/© 2024 The Authors. Published by Elsevier Inc. This is an open access article under the CC BY license ( http://creativecommons.org/licenses/by/4.0/ ). 1. Introduction The World Health Organization (WHO) defines healthy ageing as “the process of developing and maintaining the functional ability that enables well-being in older age”(Beard et al., 2016). Functional ability is determined by the intrinsic capacity (IC) of individuals and by the environment they live in. IC is therefore crucial for understanding and promoting healthy ageing. However, developing measures for longitudinal research on IC presents various hurdles. One major challenge in developing measures of IC lies in the complexities of its measurement model. The current consensus identifies five pivotal domains—locomotion, cognition, vitality, psychology, and sensory—as key components of IC, which refers to the composite of physical and mental capacities an individual can draw on (Koivunen et al., 2022;George et al., 2021;World Health Organization, 2017). These five domains have been used in WHO's reports on healthy ageing and have also been employed in various other studies concerning IC ((World Health Organization, 2019a;Ma et al., 2020;Rodríguez-Laso et al., 2023;Yan Wang et al., 2024). Nonetheless, it is important to note that research on IC is still evolving, and the identification of these five domains might not be definitive. IC transforms the idea of “healthy ageing”from a disease-focused to a function-focused perspective. Researchers have increasingly recognized the value of IC in measuring individual capacities and its connection with various aspects of functional ability in the context of ageing. The construct has been successfully validated and empirically examined in different cohorts (Beard et al., 2019;Beard et al., 2022;Si et al., 2023). Yet, the measurement model that guides the operationalization of IC in practice has not been explicitly defined in these studies. In a recent scoping review by our group, it has been suggested that IC should be examined as a formative construct (Koivunen et al., 2022). In a formative measurement model, the construct is formed by the combination of its domains, each representing a different facet that collectively defines the construct (Bollen and Diamantopoulos, 2017). Specifically, the formative approach means that individuals' total IC is constituted by their capacities regarding vitality, sensory, cognition, psychology and locomotion (Koivunen et al., 2022;Koivunen et al., 2023;Cesari et al., 2018). As a result, each domain contributes to the overall construct of IC, and changes in any domain can affect the total IC. This is in contrast to situations where a concept corresponds to a reflective measurement model, which means that changes in the overall construct would be reflected in changes across all individual indicators (Bollen and Diamantopoulos, 2017). Given this formative nature of IC, which encompasses five distinct domains, selecting indicators for composite measures should ideally be done at the domain-specific level to indicate the contributions of each domain (Fleuren et al., 2018). Another challenge to measures of IC has been the plethora of ways to select data for each specific domain of IC (Diaz and Banerjee, 2023). Since the introduction of IC, extensive research has been conducted to measure and validate this construct (Beyene et al., 2024;Sanchez-Niubo et al., 2020). Our group has developed a cross-sectional composite IC score using prediction modelling (Koivunen et al., 2023). The developed IC score included seven indicators covering all five domains of the IC. Recently, more guidelines have become available to better define IC, as a collection of articles have provided recommendations for items to include in standardized questionnaires and identified measurement areas in need of further research (Diaz and Banerjee, 2023). For instance, it is recommended that nutritional assessment should be considered in the vitality domain to capture the individual's capacity to maintain homeostasis (Cesari et al., 2022). These advancements in understanding IC's components suggested the importance of revisiting and refining our measurement approaches to ensure they capture the construct's full breadth and depth. Additionally, prioritising measures on capacities over (solely) deficits promotes a more holistic and positive approach to understanding healthy ageing. Establishing a reliable and consistent measure for longitudinal studies to monitoring IC is crucial, as emphasized by the ICOPE model, to effectively monitor and understand changes in individual capacities over time (World Health Organization, 2019a). Most earlier studies on development of measures on IC have been cross-sectional, with the exception of Beyene et al. (Beyene et al., 2024), where IC measure was developed and validated using longitudinal data repositories. The current study aimed to operationalize the concept of IC into standardized measures by developing scores for the five domains and an overall composite of IC across four measurement waves. We established a formative measurement model, maintaining the five domains as distinct dimensions. Indicators were identified for each domain individually, rather than assuming that all five domains stem from overall IC as in reflective models. Finally, the construct validity and longitudinal criterion validity of these scores were tested by examining their relationship with functioning over time. 2. Methods 2.1. Study sample The present study is part of the ongoing IMPROve (Intrinsic Capacity Maintenance for Promoting Healthy Ageing) study, which seeks to operationalize and enhance the application of the WHO’s framework on healthy aging in research and practice. For this, we used data from the Longitudinal Aging Study Amsterdam (LASA), which is an ongoing longitudinal study based on a nationally representative sample of the Dutch older population (Hoogendijk et al., 2020). Briefly, a random sample was drawn from population registers from eleven municipalities in the Netherlands. The LASA study started in 1992/93 consisting of 3107 participants aged 55–85 years. Since then, data are collected approximately every 3 years with a face-to-face general interview and a medical interview, which also includes performance tests in the homes of the respondents. In 2002/03, a second cohort of participants aged 55–64 years was added using the same sampling frame as the original cohort. The study was approved by the Ethical Review Board of the VU University Medical Center. All participants signed an informed consent before participating in the study. The LASA study protocol was conducted in accordance with the Declaration of Helsinki and received approval from the medical ethics committee of the VU University Medical Centre (IRB numbers: 92/138, 2002/141). For the current study, data from these two cohorts was combined. We used their measurements in respectively 1995/96 and in 2005/06 as the baseline measures for this particular study. These measurement cycles were treated as baseline measurement because not all relevant indicators for operationalizing IC were available at the first measurement cycle of these two cohorts in 1992/1993 and 2002/2003, respectively. For the first cohort, follow-up measurements were performed in 1998/ 99, 2001/02, and 2005/06. For the second cohort, follow-up measurements were performed 2008/09, 2011/12, and 2015/16, respectively. At baseline, the total number of participants for cohort 1 was 2545 and for cohort 2 was 908. With the combined cohorts, two study samples were used: an indicator selection sample and a score construction sample. The indicator selection sample included individuals who participated in both the general and medical interviews at baseline as some potential indicators were collected in the general interview and others in the medical interview (Fig. 1). This sample consisted of 2333 participants: 1509 from the first cohort and 824 from the second cohort. For the score construction sample, we used a larger sample of 3246 participants at baseline, aged 55 and over, including 2372 from the first cohort and 874 from the second cohort (see Fig. 2). This sample is larger in comparison to the indicator selection sample, as some missings on the indicators were allowed when constructing the domain-specific scores (see section IC scores construction), but not during the selection of indicators for the domains. All participants completed the general interview, but not all underwent the medical interview. Some participants who did not participate at one wave did participate in later waves. This Y. Qi et al. Experimental Gerontology 197 (2024) 112599 2 Fig. 1. Flowchart of the indicator selection sample *In 1995–1996, only people born before 1931 were asked to participate in the medical interview. Fig. 2. Flowchart of the score construction sample. Y. Qi et al. Experimental Gerontology 197 (2024) 112599 3 score construction sample was also used to evaluate whether the IC scores consistently and accurately differentiate individuals across different age groups. 2.2. Steps of IC scores development and validation Using the five-domain structure of IC, which has been employed in the ICOPE screening tool and other studies (Koivunen et al., 2022; George et al., 2021;World Health Organization, 2019a), we followed a stepwise procedure to create the domain-specific IC scores and the composite IC scores. 2.2.1. Indicator selection First, we identified candidate indicators of IC based on our past work (Koivunen et al., 2022;Koivunen et al., 2023), literature search, and expert opinions. These indicators covered the five domains of IC (Beard et al., 2019): vitality, sensory, cognition, psychology, and locomotion. Second, we verified the presence of each potential indicator in the LASA dataset to ensure the feasibility of developing longitudinal measures. A total of 22 indicators measured at baseline were considered. The inclusion of indicators under the vitality domain was guided by the working definition of vitality as proposed by the WHO working group and their consensus regarding potential markers to measure vitality capacity. Hand grip strength (George et al., 2021;Beard et al., 2019; Koivunen et al., 2023;Guti´ errez-Robledo et al., 2021;Guti´ errez-Robledo et al., 2019), peak flow (George et al., 2021;Beard et al., 2022;van Schoor et al., 2012), calf circumference (Sanchez-Rodriguez et al., 2023), appetite (Gaussens et al., 2023), sleep quality, and self-reported weight change (Gaussens et al., 2023) were considered to construct the vitality domain. These indicators cover several important attributes of vitality. Hand grip strength and peak flow measure neuromuscular function. Calf circumference measures body composition (Bautmans et al., 2022). Appetite was chosen as age-related physiological changes, such as decreased ghrelin release in the stomach, are known to increase feelings of fullness and reduce appetite (Cox et al., 2020). Self-reported weight loss assesses the nutritional aspect and sleep quality evaluates energy levels. Sensory function includes vision and hearing (George et al., 2021).Self-rated items on hearing in a conversation, being able to use a normal telephone, near vision, and far vision were considered to construct the sensory domain (Guti´ errez-Robledo et al., 2019). General cognitive functioning (L´ opez-Ortiz et al., 2022), information processing speed (Koivunen et al., 2023), and episodic memory (Koivunen et al., 2023) were considered to construct the cognition domain. These performance-based measures are commonly used as key indicators within the cognition domain (George et al., 2021). Anxiety symptoms (L´ opez-Ortiz et al., 2022)and depressive symptoms (L´ opez-Ortiz et al., 2022) have been commonly used as indicators for the psychology domain. It is also important to question whether the absence of anxiety or depressive symptoms fully captures the spectrum of psychological capacity, particularly on the positive end. Research suggests that resources related to a sense of control and the ability to mentally adapt to adversities can be preserved or even enhanced through growth, experiences, and learning during ageing (Wister and Cosco, 2020;Charles and Carstensen, 2010). Therefore, we have also included mastery (Golino et al., 2020), self-efficacy (Koivunen et al., 2023), and selfesteem (Astrone et al., 2022) as these capacities may be essential in compensating for physiological losses. Walking speed (George et al., 2021;Chen et al., 2023), chair rise test (Beard et al., 2019), standing balance (George et al., 2021), and cardigan test were considered to construct the locomotion domain. These are commonly used performance-based measures for assessing locomotion (George et al., 2021). We have also explored the potential cross-domain relevance of grip strength and walking speed, assessing whether grip strength should be considered as indicator for locomotion (George et al., 2021) and whether walking speed should be considered as indicator for vitality (Zhao et al., 2021). Currently, it is not clear how to best measure nutritional status as an indicator of vitality. Therefore, we have additionally tested the utility of Body Mass Index (BMI) and weight change as indicators. The detailed descriptions of how these variables were measured are provided in Appendix 1. Second, in the indicator selection sample, unidimensional factor analyses and Partial Least Squares Structural Equation Modelling (PLSSEM) were used to select indicators to construct domain-specific IC scores. Using the indicator selection sample, we first rescaled all the candidate indicators of IC using the percent of maximum possible (“POMP”) method (Cohen et al., 2013;Cohen et al., 1999), which was calculated by linearly transforming each participant's raw score into a percentage of the maximum total score of the measure in the sample. The rescale was stratified by sex. After rescaling, all indicators ranged from 0 (low capacity) to 100 (high capacity). To facilitate any possible direct comparison with other studies, we reported POMP units in the present study. A POMP score of 70, for example, indicates that the score is 70 % of the maximum possible value for the measurement in question. Next, separate unidimensional factor analyses were performed on each of the five domains of IC. Indicators with loading >0.40 were selected for further analysis (Clark and Watson, 2019). This threshold was established to ensure that only indicators with a substantial relationship to the underlying domain were included. The multiple unidimensional factor analyses were performed using the “psych”package with the “fa” function in R programming for statistical computing version 4.1.2 (Revelle and Revelle, 2015). Following the initial selection process, selected indicators were incorporated into a PLS-SEM model. PLS-SEM is the preferred approach when formatively specified constructs are concerned (Hair Jr et al., 2021a). The structural model estimated the relationships between latent constructs (i.e. the five IC domains), with correlations between the five domains being estimated The measurement model specifies the relationships between each domain as latent constructs and their corresponding observed indicators (Fig. 3). For this, indicator correlation weights represent each indicator's relative importance to the construct and indicator loading represents the absolute contribution of an indicator to its construct (Lohm¨ oller, 2013) (Hair Jr et al., 2021b). It is also recommended to consider the absolute contribution of a formative indicator to the construct, which is determined by the formative indicator's loading (Cenfetelli and Bassellier, 2009). In general, indicator loadings of 0.50 (Cenfetelli and Bassellier, 2009) and higher suggest the indicator makes a sufficient absolute contribution to forming the construct, even if it lacks a significant relative contribution (Hair Jr et al., 2021b). Significance of the indicator weights was based on 10,000 bootstrap samples (Hair Jr et al., 2021a). The PLS-SEM model was estimated using SEMinR package in R programming for statistical computing version 4.1.2 (Ray et al., 2021). 2.2.2. IC scores construction In the score construction sample, the domain-specific IC scores and composite IC scores were constructed by the indicators with statistically significant weights in the PLS-SEM model. At each measurement wave (T 0 through T 3 ), a mean score (domain-specific score) was first calculated for each domain using the selected indicators. We applied the general rule that the mean domain-specific score was calculated when 50 % or more of the indicators were present (Fairclough and Cella, 1996). Subsequently, the composite IC score was computed by averaging the five domain-specific scores, but only if all five domain-specific scores were present. All computed scores are reported in POMP units, ranging from 0 to 100. Descriptive statistics were calculated for both the indicator selection sample and the score construction sample. Y. Qi et al. Experimental Gerontology 197 (2024) 112599 4 2.2.3. IC scores validation First, descriptive statistics were calculated for both the indicator selection sample and the score construction sample. In the score construction sample, we tested the construct validity of both the domainspecific and composite IC scores. Additionally, we tested the criterion validity of the composite IC score in relation to functional limitations. Construct validity was tested using the known-groups' validity (Mokkink et al., 2010), this involved assessing whether the scores could effectively distinguish between groups known to differ in IC, such as different age groups. Based on the hypothesis that IC decreases with age, we compared the IC scores across these groups. For this, data that were originally structured according to measurement wave was restructured to represent each IC domain at 3-year age intervals, covering intervals ranging from 57 to 59 years to 90+years. Scatter plots were generated using ggplot2 in R to visualize the relationship between age categories and the main score of each domain. Criterion validity was assessed by examining the association between composite IC scores and functional limitations over four measurement waves using a linear mixed-effects model (LMM). Functional limitation was assessed using a scale based on six items that measured limitations in performing certain activities. Response categories ranged from 1 to 5, and were summed to create a scale ranging from 6 to 30. A higher score indicates less functional limitation (Eekhoff et al., 2019). The detailed description of the six items can be found in Appendix 1. The LMM model was formulated with functional limitation as dependent variable and IC scores as primary independent variable. Both functional limitation and IC scores were measured at four time points, from T 0 to T 3 . The model included a fixed effect of composite IC score for each wave and a random intercept to account for between-subject variability. Age and sex were included as covariates. The model was fitted using the restricted maximum likelihood (REML) estimation method to provide unbiased variance component estimates. The Wald test was used to assess the significance of fixed effects, and confidence intervals were calculated for model parameters. Model diagnostics were performed to evaluate the assumptions of homoscedasticity and normality of residuals. The LMM model was estimated using the lme4 package for R programming for statistical computing version 4.1.2 (Bates, 2010). 3. Results The average age of participants in the indicator selection sample was 71.3 years, with a standard deviation (SD) of 8.3 years. Among the participants, 51.8 % were female (Table 1). The score construction sample had a mean age of 70.0 years (SD =8.6) with 53.2 % female participants. The relatively large number of missing values for calf circumference was due to the introduction of this measurement during the year of data collection, resulting in the first participants not being measured. 3.1. Indicator selection Table 2 shows the results of the unidimensional factor analyses in which each IC domain was considered individually in relation to their corresponding candidate indicators. Based on standardized loadings from the pattern matrix, three variables—grip strength (0.62), peak flow (0.71), and calf circumference (0.48)—demonstrated loadings above the 0.40 threshold, indicating a significant contribution to the vitality domain. For sensory, cognition and psychology, all candidate indicators showed loadings higher than the threshold. For locomotion, only the balance test showed a loading below the threshold. Based on these results, a total of 18 candidate (in bold) indicators with loading above 0.40 were selected to fit in the PLS-SEM model. 3.2. Indicator selection-additional analyses We have additionally evaluated the utility of Body Mass Index (BMI) alongside weight change as indicators of vitality by assigning scores that reflect increasing levels of capacity. Specifically, BMI was categorized from 1 to 4, with 1 representing underweight (BMI <18.5), 2 indicating obesity (BMI higher than 30), 3 for overweight (BMI between 25 and 29.9), and 4 corresponding to normal weight (BMI between 18.5 and 24.9). Weight change was similarly scored: 1 for involuntary weight loss, 2 for weight gain due to any reason, 3 for voluntary weight loss, and 4 for no changes in weight. Neither of these variables demonstrated acceptable correlations with vitality. In unidimensional factor analysis, BMI correlated at 0.12 after adjustment for overlaps with other indicators, falling well below our threshold of 0.40. Weight change correlated at only 0.05 after adjustment for overlaps with other indicators, also falling below the threshold. Table 3 shows the indicator weights and their corresponding 95 % bootstrap confidence intervals for each IC domain. If a confidence Table 1 Sample characteristics of the indicator selection sample (n=2333) and the score construction sample (n=3246) at T 0 , as Percentage of Possible Maximum (POMP) units. Indicator selection sample Score construction sample Mean (SD)/ Percentage N. valid cases Mean (SD)/ Percentage N. valid cases Age 71.3(8.3) 2333 70.0(8.6) 3246 Sex(female) 51.8 % 2333 53.2 % 3246 Vitality Hand grip strength 53.1 (14.8) 2279 53.1 (14.8) 2279 Peak flow 49.8 (17.6) 2115 49.8 (17.6) 2130 Calf circumference 42.3 (17.3) 1547 42.0 (16.8) 1972 Appetite 93.6 (19.2) 2286 93.8 (18.8) 3084 Sleep quality 67.9 (23.6) 2185 68.5 (23.4) 2844 Weight change 75.1 (36.2) 2322 75.1 (36.2) 2322 Sensory Hearing in a conversation 83.5 (27.4) 2326 84.9 (26.3) 3115 Use normal telephone 96.3 (15.9) 2326 96.6 (15.3) 3115 Near vision 88.3 (22.8) 2331 88.0 (22.9) 3122 Far vision 93.2 (18.9) 2324 93.2 (19.0) 3115 Cognition General cognitive functioning 87.1 (13.3) 2329 89.2 (11.6) 3153 Information processing speed 48.2 (18.0) 2217 48.2 (18.0) 2217 Episodic memory 56.5 (17.9) 2283 56.5 (17.9) 2283 Psychology Anxiety symptoms 85.0 (17.4) 2284 85.8 (16.7) 3080 Depressive symptoms 83.0 (15.8) 2277 84.2 (15.3) 3072 Mastery 60.5 (17.9) 2243 62.9 (17.1) 3025 Self-efficacy 55.4 (13.8) 2257 55.5 (13.9) 3040 Self esteem 67.6 (15.4) 2263 68.9 (14.9) 3047 Locomotion Walking speed 87.9 (7.3) 2238 88.1 (7.2) 3015 Chair rise test 81.1 (10.7) 2103 84.9 (7.3) 2823 Balance test 79.1 (38.5) 2248 80.6 (37.3) 3020 Cardigan test 84.7 (9.3) 2281 85.0 (9.2) 3080 Y. Qi et al. Experimental Gerontology 197 (2024) 112599 5 interval does not include the value zero, the weight can be considered statistically significant, and the indicator can be retained (Hair Jr et al., 2021a). The analysis of indicator weights concludes the evaluation of the formative measurement models. In the current model, all indicators showed significant weights corresponding to their domains, therefore all indicators were retained (Cenfetelli and Bassellier, 2009). Fig. 3 visualizes the structural model together with indicator loadings of each IC indicator. For the structural model, bivariate correlations were estimated in regard to the relationship between the five domains. The five domains did not exhibit high correlations between each other (range 0.03 to 0.46). For the current model, only calf circumference showed loading that was lower than 0.50, with other loadings ranging from 0.60 to 0.88. However, as the weight of calf circumference was statistically significant, we have retained calf circumference as indicator for vitality. 3.3. IC scores validation We computed five domain-specific IC scores and one composite IC score for each measurement wave, using the statistically significant indicators from the PLS-SEM model. Descriptive statistics, baseline characteristics, IC domain scores and composite scores for the score construction sample at each measurement wave are available in Appendix 2. Construct validity was assessed using the known-groups' validity by testing whether the scores could effectively distinguish between different age groups. Fig. 4 presents the domain-specific IC scores and composite IC scores across different age groups. Each point represents the mean score for a specific age group of each domain-specific score. The x-axis represents the age groups, and the y-axis depicts the mean domain-specific scores. The vertical lines extending above and below each mean score represent the standard deviation above and below the mean score for each age group. Overall, we observed a clear trend of lower IC with increasing age across all domains. The average vitality score was lower in higher age groups, starting at an average of 58 (SD 10) for the youngest group and dropping to 36 (SD 11) for those aged 90 or older. The average sensory and locomotion scores also followed a consistent decline up to the 90 years or older group, with increased variability in these scores observed with age. The cognition and psychology domain both showed similar trends of decline with ageing. The variability in the cognition scores remained rather consistent in comparison to other domains. The composite IC score also showed a steady decline from 78 to 60, with relatively small variability. The mixed effects linear regression model with functional limitations as outcome achieved a R 2 of 0.76. Fixed effects revealed significant predictors including composite IC score (B =0.20, p<0.001), age (B = -0.10, p <0.001), and sex (B =-1.32, p <0.001), each statistically significant with Satterthwaite degrees of freedom adjustments. Random effects showed substantial variability attributed to individual respondents (ICC =0.67) and minimal variability across the four measurement waves (ICC =0.01). Overall, the mixed effects linear regression model indicates that higher IC scores were related to less functional limitations, as evidenced by the upward slope of the regression line. The figure showing the relationship between composite IC score and functional limitation can be found in Appendix 4. 4. Discussion The current study aimed to operationalize the concept of IC into standardized longitudinal measures by adopting a formative measurement model and utilizing data from the Longitudinal Aging Study Table 2 Factor loadings of candidate IC indicators: separate unidimensional factor analysis. Vitality Sensory Cognition Psychology Locomotion Hand grip strength 0.62 Peak flow 0.71 Calf circumference 0.48 Appetite 0.28 Sleep quality 0.19 Weight change 0.05 Hearing in a conversation 0.52 Use normal telephone 0.53 Near vision 0.50 Far vision 0.43 General cognitive functioning 0.68 Information processing speed 0.77 Episodic memory 0.70 Anxiety symptoms 0.65 Depressive symptoms 0.75 Mastery 0.72 Self-efficacy 0.57 Self esteem 0.66 Walking speed 0.78 Chair rise test 0.68 Balance test 0.19 Cardigan test 0.62 Table 3 Measurement model of the five IC domains: Indicator weights and bootstrap 95 % confidence interval, relative importance of one indicator to its domain based on results from PLS-SEM model. Vitality Sensory Cognition Psychology Locomotion Hand grip strength 0.65 [0.62, 0.69] Peak flow 0.50 [0.47, 0.53] Calf circumference 0.08 [0.02, 0.14] Hearing in a conversation 0.44 [0.40, 0.49] Use normal telephone 0.32 [0.27, 0.37] Near vision 0.41 [0.37, 0.46] Far vision 0.33 [0.28, 0.38] General cognitive functioning 0.39 [0.36, 0.42] Information processing speed 0.45 [0.43, 0.48] Episodic memory 0.41 [0.39, 0.43] Anxiety symptoms 0.10 [0.06, 0.14] Depressive symptoms 0.34 [0.30, 0.37] Mastery 0.35 [0.32, 0.39] Self-efficacy 0.37 [0.33, 0.40] Self esteem 0.16 [0.12, 0.19] Walking speed 0.61 [0.55, 0.66] Chair rise test 0.26 [0.21, 0.31] Cardigan test 0.38 [0.33, 0.43] Y. Qi et al. Experimental Gerontology 197 (2024) 112599 6 Amsterdam. We have identified 18 indicators covering the five domains of IC. The developed domain-specific IC scores and composite IC score demonstrated good validity and indicated age-related declines. Higher composite IC scores were associated with fewer functional limitations over time. Guided by the WHO's definition of IC, our study used the five Fig. 3. Structural model of the five domains of IC with results of the path analysis. Fig. 4. Age group comparison in domain-specific IC scores and composite IC scores to test known groups' validity, as Percentage of Possible Maximum (POMP) units. Y. Qi et al. Experimental Gerontology 197 (2024) 112599 7 domains that are commonly recognized in the literature (World Health Organization, 2019a;Rodríguez-Laso et al., 2023;World Health Organization, 2019b). Our objective was not to explore the conceptualisation of IC, but rather to operationalize the domains of IC using existing data sources and framework. Building on our previous work, we assumed a formative measurement model for IC. The formative approach is rarely acknowledged in the development of IC measures, even though IC is recognized as a multidimensional construct (Koivunen et al., 2023; Cesari et al., 2018;Nascimento et al., 2023) that could be more appropriately operationalized with composite measures rather than reflective scales. We demonstrated an approach for developing both domainspecific and composite IC measures using formative constructs. A formative model suggests that IC is an operationalization of a multidimensional construct, summarizing various conceptually distinct domains (World Health Organization, 2019a). This operationalization aligns with the findings from the network analysis by Koivunen et al., which indicates that IC should not be conceptualized as stemming from a single, general trait (Lohm¨ oller, 2013). The recognition of IC as a formative construct has important implications for research. It necessitates the use of appropriate statistical techniques that can accurately capture the different dimensions of this construct, moving beyond techniques that assume an overall latent variable model. This perspective also influences how interventions are designed and implemented. Since IC is not derived from a single trait but from multiple independent domains, interventions can be more precisely tailored to target specific areas of deficiency within an individual's intrinsic capacity profile, focusing on achieving specific outcomes. At the domain level, we considered a total of 22 candidate indicators, which spanned the five IC domains. While the selection of candidate indicators for the sensory, cognition, and psychology domains was relatively straightforward, identifying appropriate indicators for the vitality and locomotion domains proved more challenging. The current working definition of vitality suggests that it represents a physiological state, arising from either normal or accelerated biological ageing processes (Bautmans et al., 2022). This state is the outcome of interactions among multiple physiological systems and is manifested in various bodily functions such as energy and metabolism, neuromuscular function, and stress response capabilities (Beard et al., 2019;Bautmans et al., 2022). We have considered hand grip strength, peak flow, calf circumference, appetite, sleep quality, and weight as indicators for vitality. Among these, hand grip strength, peak flow, and calf circumference were selected to construct the domain score for vitality. Hand grip strength is often used to measure vitality (Beard et al., 2019;Beard et al., 2022;Aliberti et al., 2022) as it captures the vital sign of physiological reserve and biological age (Granic et al., 2017;Sayer and Kirkwood, 2015;Lu et al., 2023). Peak flow, similarly, has been utilized in studies as an indicator of vitality (Guti´ errez-Robledo et al., 2019;van Schoor et al., 2024). Although not as commonly employed, calf circumference has demonstrated its relevance as a significant marker of nutritional status (Bonnefoy et al., 2002;Cruz-Jentoft et al., 2018). Together, these three indicators can possibly provide a comprehensive measure of vitality. Among the indicators not selected for the vitality domain, weight change exhibited the lowest loading. We evaluated the utility of BMI and weight change as indicators of vitality, but none of these variables were selected as indicator due to non-significant correlations and weights. This finding suggests that, within our sample, BMI and weight change may not effectively capture the essence of vitality, particularly in relation to metabolism. This could be due to these indicators not adequately reflecting the same physiological and metabolic components of vitality or that the operational definitions used did not capture the necessary aspects of these measures. This finding suggests the need for further exploration into measures that might more accurately reflect the metabolism and stress response aspects of the vitality domain. The broader issue highlighted by our results is that the operational definitions of vitality domain is still being developed (Bautmans et al., 2022). The empirical meaning of a construct might differ from its intended meaning based on the chosen indicators. This risk exists for both formative models and reflective models and ties into the issue of indicator validity (Fleuren et al., 2018). We may also need to extend this consideration to many other constructs, which should always be considered when interpreting research findings. Hand grip strength has commonly been used as an indicator for vitality, reflecting overall muscle strength and physiological resilience. However, given muscle strength's importance in activities requiring bodily stabilization and support, hand grip strength might also be pertinent to locomotion (Nayasista et al., 2022). The additional unidimensional factor analyses (Appendix 3) showed that hand grip strength had a loading of 0.46 in the locomotion domain, which is lower than its loading in the vitality domain (0.62), reaffirming its primary association with vitality. Additionally, walking speed, which results from the interplay of an individual's physiological capabilities and their perceptions of environmental conditions and task demands, has been advocated as sign of vitality (Middleton et al., 2015). Our analyses found that walking speed had a loading of 0.27 in the vitality domain, which is lower than its loading in the locomotion domain. This finding confirms walking speed as indicator for the locomotion domain within the IC framework. Finally, we assessed the construct validity of both the developed domain-specific IC scores and the composite IC scores. Unlike reflective models, where factor analysis can confirm construct validity, formative measurement models do not derive meaningful interpretive value from statistically calculating internal consistency, nor do they confirm a single underlying latent variable (Fleuren et al., 2018). This is due to formative measurement models not assuming unidimensionality and, as such, requiring more intricate methods for assessing validity, such as construct and criterion validity (Fleuren et al., 2018). Our findings indicate a consistent decline with age across the individual domains of vitality, sensory, locomotion, cognition, and psychology, as well as in the overall IC score. This uniform decline suggests the robust construct validity of our IC scores. Collectively, these results validate that the IC scores effectively mirror the gradual decline of IC, a fundamental aspect of ageing. It also suggests that focusing only on the aggregated overall IC score across multiple domains might lead to the loss of information about the different contributing aspects (Koivunen et al., 2023). While our composite score is a comprehensive representation of IC and has practical advantages, the multidimensional nature of the composite IC score also represents a potential limitation. It is for example possible that individual domains will change independently of each other. These changes might be “hidden”from the researchers view if only the composite IC is examined. Therefore, we recommend that researchers always compute and report both domain-level scores along with the composite IC scores. It is worth mentioning that most studies, including ours, have used functional abilities-related measures to validate the developed IC measure (Beard et al., 2019;Beyene et al., 2024;Guti´ errez-Robledo et al., 2021;Salinas-Rodríguez et al., 2022). The WHO's healthy ageing model suggests that an individual's level of IC significantly influences their functional ability in interaction with the surrounding environment (Belloni and Cesari, 2019). Following this idea, we considered functional decline as a likely result of reduced IC and used it to validate the developed composite IC score. However, IC is a unique construct and thus one suitable gold standard does not exist yet (Hoogendijk et al., 2023). We used POMP scores to construct each domain-specific score and subsequently averaged these to obtain the composite IC score. POMP scores are akin to summed scores in that they are not derived from a Y. Qi et al. Experimental Gerontology 197 (2024) 112599 8 World Health Organization, 2021. Decade of Healthy Ageing: Baseline Report. World Health Organization. Yan Wang, N., Liu, X., Kong, X., Sumi, Y., Chhetri, J.K., Hu, L., et al., 2024. Implementation and impact of the World Health Organization integrated care for older people (ICOPE) program in China: a randomised controlled trial. Age Ageing 53(1):afad249. Zhao, J., Chhetri, J.K., Chang, Y., Zheng, Z., Ma, L., Chan, P., 2021. Intrinsic capacity vs. multimorbidity: a function-centered construct predicts disability better than a disease-based approach in a community-dwelling older population cohort. Front. Med. 8, 753295. Y. Qi et al. Experimental Gerontology 197 (2024) 112599 15